Fix scf.for loop-carried memref reconciliation

[HecreReed]

Summary

• reconcile scf.for loop-carried iter_arg/result types after PTOViewToMemref rewrites tile values to memrefs
• add a reduced .pto regression for loop-carried tile_buf values
• verify the reported qwen3_decode_layer_incore_0.pto now compiles on top of origin/main

Root cause

PTOViewToMemref already had a post-rewrite fixup for scf.if, but not for scf.for. When loop-carried tile_buf values were rewritten to memrefs inside the loop body, the pass left the scf.for init operand, region iter_arg, yield operand, and op result out of sync.

Verification

• cmake --build /tmp/ptoas-qwen-repro.oy3rK0/build-fix --target ptoas -j8
• ptoas /tmp/ptoas-qwen-repro.oy3rK0/fix-branch/test/basic/scf_for_tile_iter_arg_reconcile.pto 2>&1 1>/dev/null | FileCheck /tmp/ptoas-qwen-repro.oy3rK0/fix-branch/test/basic/scf_for_tile_iter_arg_reconcile.pto
• ptoas /Users/laoda/Downloads/qwen3_decode_layer_incore_0.pto -o /tmp/qwen-fix.cpp